kantver - Fotolia

What mistakes throw off VM load balancing?

I track VM performance on my IT infrastructure, and want to optimize the distribution of virtual workloads.

Planning VM capacity based on actual IT infrastructure utilization fully exploits virtualization technology and allows for less spending. To mitigate the risk of exhausting IT resources and get the most of a company's infrastructure, administrators can turn to VM load balancing automation to optimize their workloads.

Good virtualization analytics enables IT teams to make decisions and carry out actions based on infrastructure utilization and performance data. With analysis of VM behavior and workloads, IT admins can automate how the virtualization deployment uses IT resources. Generally, automate:

  • Workload initial placement and replacement, to balance the load evenly across the infrastructure at all times; and
  • Workload right-sizing, scaling up or down, to identify and fix when a VM is oversized or undersized.

Using an automation engine with virtual machines is an extension of business logic: Start with decisions, based on analytics, about what a positive result looks like. Based on those decisions, create actions for the automation engine to take based on information from the analytics system.

To optimize a VM deployment, start by establishing good analytical data, with an understanding of workload needs and long-term requirements.

To automate VM load balancing, avoid looping actions, forecast all results of an action and consider all of the workload's key metrics when taking an action upon it.

Looping actions. If the administrator moves a VM workload from Host1 to Host2, the inverse decision should not be made on the following analytical pass. It would create a ping-pong effect of the VM bouncing between these two hosts on every scheduled run. This looping action could actually cause more problems than it is trying to solve.

The state of the environment post-action. If an action ends up resulting in a worse situation than it attempts to abate, then the automation and analytics engine is not doing its job. Good decision-making tools have a predictive component built in, and VM administrators and architects should have a predictive mindset when choosing actions.

Key metrics for a workload. There's no point in taking action to ease contention on one resource if it creates contention on another. Or maybe there is! Perhaps a memory-bound VM would do better to have less memory pressure but be capped a little in terms of CPU. There's not always a perfect answer and incremental progress may be acceptable in some scenarios.

The payoffs to automatically load balancing virtual machines include lower spending on data center IT equipment and power, but also higher IT uptime thanks to the agile responsiveness of the virtualized infrastructure. VM load balancing also reduces performance related issues, which translates to more productivity in the organization as a whole. When automation and analytics work together, IT can better predict the growth and future needs of its infrastructure, giving CFOs better visibility into future spending.

Next Steps

Load balancing your VMs with dynamic optimization

Should I choose a software or hardware load balancer?

Load balancing meets university's apps needs

Digging deeper into VMware DRS and load balancing

Dig Deeper on Matching IT Resources to Application Requirements